In January, Claude Code became popular and sparked debates about the end of SaaS, with software-related stocks erasing about $300 billion in a single day. Tech writer Joan Westenberg then responded with the bread paradox: Americans buy about 10 million pre-baked bread loaves every day—despite bread machines costing less than $100 and recipes having been around for thousands of years—because the hidden costs of making it at home determine that demand for external purchases continues, and enterprise procurement of SaaS follows the same logic.
The historical foundation of the bread paradox: the persistence of a five-thousand-year commercial baking supply chain
In her essay, Joan Westenberg traces the history of commercial baking back to 3,000 BCE: the ancient Egyptians were already running commercial bakeries along the Nile; in the Roman era, professional breadmakers’ guilds created specialization, animal power, and baking ovens plus a city distribution system, allowing hundreds of thousands of residents to avoid baking themselves. By the 12th century, London had a royal chartered bakers’ guild, in 1928 Otto Rohwedder invented the commercial slicing machine, and in 1961 the Chorleywood process compressed baking time to a few minutes.
Today the United States consumes about 21 million tons of bread products every year, and still buys about 10 million pre-baked bread loaves per day. Westenberg points out that, using the economics framework of “make-or-buy,” most people underestimate the hidden costs of “making it yourself”—ingredients, operation, waiting, and cleaning. Each step seems small on its own, but when repeated over the long term it creates a clear cost gap, which is the core reason home production cannot eliminate the commercial baking industry.
AI code security costs and the current state of SaaS procurement
When companies use AI to build systems in-house, they face quantifiable hidden costs. According to relevant research, the number of major defects in AI-generated code is about 1.7 times that of code written by humans, and the risk of security gaps is higher than with human-developed code. The original article also lists, as specific risks, that after key personnel leave, in-house systems can lose readability and there may be no one to maintain them.
In contrast to the real dynamics of the SaaS procurement market, Gartner has observed that recent SaaS renewals typically see upsides in the range of 10% to 20%, outpacing the budget growth rate of most CIOs—but buyers are not leaving in large numbers. A report published by Avenir in January 2026 shows that 63% of enterprise software procurers expect existing software vendors to benefit from generative AI, while only 8% believe vendors will be harmed. All of the above data point to higher market confidence in how SaaS vendors will evolve with AI than assessments of substitution threats.
The Klarna case and the thin product boundary: the actual approach to enterprise SaaS replacement
Klarna, frequently cited as an “AI in-house build beats SaaS” example, actually uses a different SaaS portfolio to replace Salesforce, not building a complete replacement system from scratch using AI. Klarna’s teams still use Slack under Salesforce to this day, showing that enterprise SaaS replacement is mostly about horizontal platform switching rather than exiting the SaaS ecosystem.
In her argument, Westenberg distinguishes two categories of SaaS products: SaaS platforms with deep integration, proprietary data, regulatory certifications, years of business logic, and an ecosystem of partners. This corresponds to the industrial complex of commercial baking—the supply-chain logic does not change just because AI lowers code costs. By contrast, thin products where the subscription charge hinges on a single function that can be copied with one AI prompt (such as PDF conversion or automated meeting notes). Westenberg says their rationale depends on a business environment where software development is expensive in the first place, which is a different category from deeply integrated SaaS platforms.
Frequently asked questions
What is the core of the “bread paradox” argument?
Tech writer Joan Westenberg uses the “bread paradox” to explain that even though bread machines cost less than $100 and recipes have circulated for five thousand years, Americans still buy about 10 million pre-baked bread loaves every day because “the hidden costs of making it at home” are seriously underestimated. She draws a parallel to enterprise SaaS procurement logic: the real costs of using AI to build in-house systems (maintenance, security flaws, and loss of readability after personnel turnover) are often underestimated.
Are there specific data on the security risks of AI-generated code?
According to relevant research, the number of major defects in AI-generated code is about 1.7 times that of code written by humans. Gartner has observed that even after AI code becomes widespread, enterprise SaaS renewal growth still lands in the range of 10% to 20%. Avenir’s January 2026 report also shows that 63% of enterprise software procurers expect SaaS vendors to benefit from AI.
What is Klarna’s actual strategy for “replacing SaaS”?
Per reports, Klarna’s actual approach is to replace Salesforce with another suite of SaaS, not to use AI to build a complete system from scratch. Its teams still use Slack under Salesforce to this day—this is a horizontal platform replacement rather than exiting the SaaS ecosystem.
